Disable flake LFS

Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>

.gitattributes

@@ -0,0 +1 @@
+*.pdf filter=lfs diff=lfs merge=lfs -text

flake.nix

@@ -5,6 +5,7 @@
     agenix.inputs.nixpkgs.follows = "nixpkgs";
     bscpkgs.url = "git+https://git.sr.ht/~rodarima/bscpkgs";
     bscpkgs.inputs.nixpkgs.follows = "nixpkgs";
+    self.lfs = false;
   };
 
   outputs = { self, nixpkgs, agenix, bscpkgs, ... }:

keys.nix

@@ -16,8 +16,7 @@ rec {
   };
 
   hostGroup = with hosts; rec {
-    untrusted  = [ fox ];
+    compute    = [ owl1 owl2 fox ];
-    compute    = [ owl1 owl2 ];
     playground = [ eudy koro weasel ];
     storage    = [ bay lake2 ];
     monitor    = [ hut ];
@@ -31,6 +30,7 @@ rec {
   admins = {
     "rarias@hut"  = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIE1oZTPtlEXdGt0Ak+upeCIiBdaDQtcmuWoTUCVuSVIR rarias@hut";
     "rarias@tent" = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIIwlWSBTZi74WTz5xn6gBvTmCoVltmtIAeM3RMmkh4QZ rarias@tent";
+    "rarias@fox"  = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIDSbw3REAKECV7E2c/e2XJITudJQWq2qDSe2N1JHqHZd rarias@fox";
     root          = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIII/1TNArcwA6D47mgW4TArwlxQRpwmIGiZDysah40Gb root@hut";
   };
 }

m/apex/configuration.nix

@@ -5,7 +5,9 @@
     ../common/xeon.nix
     ../common/ssf/hosts.nix
     ../module/ceph.nix
+    ../module/slurm-server.nix
     ./nfs.nix
+    ./wireguard.nix
   ];
 
   # Don't install grub MBR for now

m/apex/nfs.nix

@@ -8,6 +8,7 @@
     statdPort = 4000;
     exports = ''
       /home 10.0.40.0/24(rw,async,no_subtree_check,no_root_squash)
+      /home 10.106.0.0/24(rw,async,no_subtree_check,no_root_squash)
     '';
   };
   networking.firewall = {
@@ -27,6 +28,21 @@
       iptables -A nixos-fw -p udp -s 10.0.40.0/24 --dport 4001  -j nixos-fw-accept
       iptables -A nixos-fw -p udp -s 10.0.40.0/24 --dport 4002  -j nixos-fw-accept
       iptables -A nixos-fw -p udp -s 10.0.40.0/24 --dport 20048 -j nixos-fw-accept
+
+      # Accept NFS traffic from wg0
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 111   -j nixos-fw-accept
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 2049  -j nixos-fw-accept
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 4000  -j nixos-fw-accept
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 4001  -j nixos-fw-accept
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 4002  -j nixos-fw-accept
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.0/24 --dport 20048 -j nixos-fw-accept
+      # Same but UDP
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 111   -j nixos-fw-accept
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 2049  -j nixos-fw-accept
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 4000  -j nixos-fw-accept
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 4001  -j nixos-fw-accept
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 4002  -j nixos-fw-accept
+      iptables -A nixos-fw -p udp -i wg0 -s 10.106.0.0/24 --dport 20048 -j nixos-fw-accept
     '';
   };
 }

m/apex/wireguard.nix

@@ -0,0 +1,35 @@
+{ config, ... }:
+
+{
+  networking.firewall = {
+    allowedUDPPorts = [ 666 ];
+  };
+
+  age.secrets.wgApex.file = ../../secrets/wg-apex.age;
+
+  # Enable WireGuard
+  networking.wireguard.enable = true;
+  networking.wireguard.interfaces = {
+    # "wg0" is the network interface name. You can name the interface arbitrarily.
+    wg0 = {
+      ips = [ "10.106.0.30/24" ];
+      listenPort = 666;
+      privateKeyFile = config.age.secrets.wgApex.path;
+      # Public key: VwhcN8vSOzdJEotQTpmPHBC52x3Hbv1lkFIyKubrnUA=
+      peers = [
+        {
+          name = "Fox";
+          publicKey = "VfMPBQLQTKeyXJSwv8wBhc6OV0j2qAxUpX3kLHunK2Y=";
+          allowedIPs = [ "10.106.0.0/24" ];
+          endpoint = "fox.ac.upc.edu:666";
+          # Send keepalives every 25 seconds. Important to keep NAT tables alive.
+          persistentKeepalive = 25;
+        }
+      ];
+    };
+  };
+
+  networking.hosts = {
+    "10.106.0.1" = [ "fox" ];
+  };
+}

m/common/base.nix

@@ -3,6 +3,7 @@
   # Includes the basic configuration for an Intel server.
   imports = [
     ./base/agenix.nix
+    ./base/always-power-on.nix
     ./base/august-shutdown.nix
     ./base/boot.nix
     ./base/env.nix

m/common/base/always-power-on.nix

@@ -0,0 +1,8 @@
+{
+  imports = [
+    ../../module/power-policy.nix
+  ];
+
+  # Turn on as soon as we have power
+  power.policy = "always-on";
+}

m/common/base/august-shutdown.nix

@@ -1,12 +1,12 @@
 {
-  # Shutdown all machines on August 2nd at 11:00 AM, so we can protect the
+  # Shutdown all machines on August 3rd at 22:00, so we can protect the
   # hardware from spurious electrical peaks on the yearly electrical cut for
   # manteinance that starts on August 4th.
   systemd.timers.august-shutdown = {
-    description = "Shutdown on August 2nd for maintenance";
+    description = "Shutdown on August 3rd for maintenance";
     wantedBy = [ "timers.target" ];
     timerConfig = {
-      OnCalendar = "*-08-02 11:00:00";
+      OnCalendar = "*-08-03 22:00:00";
       RandomizedDelaySec = "10min";
       Unit = "systemd-poweroff.service";
     };

m/common/base/env.nix

@@ -4,7 +4,7 @@
   environment.systemPackages = with pkgs; [
     vim wget git htop tmux pciutils tcpdump ripgrep nix-index nixos-option
     nix-diff ipmitool freeipmi ethtool lm_sensors cmake gnumake file tree
-    ncdu config.boot.kernelPackages.perf ldns pv
+    ncdu config.boot.kernelPackages.perf ldns pv git-lfs
     # From bsckgs overlay
     osumb
   ];

m/common/base/net.nix

@@ -14,7 +14,7 @@
     nftables.enable = lib.mkForce false;
 
     hosts = {
-      "84.88.53.236" = [ "apex" "ssfhead.bsc.es" "ssfhead" ];
+      "84.88.53.236" = [ "ssfhead.bsc.es" "ssfhead" ];
       "84.88.51.152" = [ "raccoon" ];
       "84.88.51.142" = [ "raccoon-ipmi" ];
     };

m/common/base/users.nix

@@ -154,6 +154,20 @@
           "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIIV5LEAII5rfe1hYqDYIIrhb1gOw7RcS1p2mhOTqG+zc pedro@pedro-ThinkPad-P14s-Gen-2a"
         ];
       };
+
+      csiringo = {
+        # Arbitrary UID but large so it doesn't collide with other users on ssfhead.
+        uid = 9653;
+        isNormalUser = true;
+        home = "/home/Computational/csiringo";
+        description = "Cesare Siringo";
+        group = "Computational";
+        hosts = [ "apex" "weasel" ];
+        hashedPassword = "$6$0IsZlju8jFukLlAw$VKm0FUXbS.mVmPm3rcJeizTNU4IM5Nmmy21BvzFL.cQwvlGwFI1YWRQm6gsbd4nbg47mPDvYkr/ar0SlgF6GO1";
+        openssh.authorizedKeys.keys = [
+          "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIHA65zvvG50iuFEMf+guRwZB65jlGXfGLF4HO+THFaed csiringo@bsc.es"
+        ];
+      };
     };
 
     groups = {

m/fox/configuration.nix

@@ -5,8 +5,15 @@
     ../common/base.nix
     ../common/xeon/console.nix
     ../module/emulation.nix
+    ../module/nvidia.nix
+    ../module/slurm-client.nix
+    ./wireguard.nix
   ];
 
+  # Don't turn off on August as UPC has different dates.
+  # Fox works fine on power cuts.
+  systemd.timers.august-shutdown.enable = false;
+
   # Select the this using the ID to avoid mismatches
   boot.loader.grub.device = "/dev/disk/by-id/wwn-0x500a07514b0c1103";
 
@@ -30,6 +37,18 @@
 
   services.openssh.settings.X11Forwarding = true;
 
+  services.fail2ban.enable = true;
+
+  # Use SSH tunnel to reach internal hosts
+  programs.ssh.extraConfig = ''
+    Host bscpm04.bsc.es gitlab-internal.bsc.es tent
+      ProxyJump raccoon
+    Host raccoon
+      ProxyJump apex
+      HostName 127.0.0.1
+      Port 22022
+  '';
+
   networking = {
     timeServers = [ "ntp1.upc.edu" "ntp2.upc.edu" ];
     hostName = "fox";
@@ -53,17 +72,20 @@
     extra-trusted-public-keys = [ "jungle.bsc.es:pEc7MlAT0HEwLQYPtpkPLwRsGf80ZI26aj29zMw/HH0=" ];
   };
 
-  # Configure Nvidia driver to use with CUDA
+  # Recommended for new graphics cards
-  hardware.nvidia.package = config.boot.kernelPackages.nvidiaPackages.production;
   hardware.nvidia.open = true;
-  hardware.graphics.enable = true;
-  nixpkgs.config.nvidia.acceptLicense = true;
-  services.xserver.videoDrivers = [ "nvidia" ];
 
   # Mount NVME disks
   fileSystems."/nvme0" = { device = "/dev/disk/by-label/nvme0"; fsType = "ext4"; };
   fileSystems."/nvme1" = { device = "/dev/disk/by-label/nvme1"; fsType = "ext4"; };
 
+  # Mount the NFS home
+  fileSystems."/nfs/home" = {
+    device = "10.106.0.30:/home";
+    fsType = "nfs";
+    options = [ "nfsvers=3" "rsize=1024" "wsize=1024" "cto" "nofail" ];
+  };
+
   # Make a /nvme{0,1}/$USER directory for each user.
   systemd.services.create-nvme-dirs = let
     # Take only normal users in fox
@@ -80,4 +102,20 @@
     wantedBy = [ "multi-user.target" ];
     serviceConfig.ExecStart = script;
   };
+
+  # Only allow SSH connections from users who have a SLURM allocation
+  # See: https://slurm.schedmd.com/pam_slurm_adopt.html
+  security.pam.services.sshd.rules.account.slurm = {
+    control = "required";
+    enable = true;
+    modulePath = "${pkgs.slurm}/lib/security/pam_slurm_adopt.so";
+    args = [ "log_level=debug5" ];
+    order = 999999; # Make it last one
+  };
+
+  # Disable systemd session (pam_systemd.so) as it will conflict with the
+  # pam_slurm_adopt.so module. What happens is that the shell is first adopted
+  # into the slurmstepd task and then into the systemd session, which is not
+  # what we want, otherwise it will linger even if all jobs are gone.
+  security.pam.services.sshd.startSession = lib.mkForce false;
 }

m/fox/wireguard.nix

@@ -0,0 +1,46 @@
+{ config, ... }:
+
+{
+  networking.firewall = {
+    allowedUDPPorts = [ 666 ];
+  };
+
+  age.secrets.wgFox.file = ../../secrets/wg-fox.age;
+
+  networking.wireguard.enable = true;
+  networking.wireguard.interfaces = {
+    # "wg0" is the network interface name. You can name the interface arbitrarily.
+    wg0 = {
+      # Determines the IP address and subnet of the server's end of the tunnel interface.
+      ips = [ "10.106.0.1/24" ];
+
+      # The port that WireGuard listens to. Must be accessible by the client.
+      listenPort = 666;
+
+      # Path to the private key file.
+      privateKeyFile = config.age.secrets.wgFox.path;
+      # Public key: VfMPBQLQTKeyXJSwv8wBhc6OV0j2qAxUpX3kLHunK2Y=
+
+      peers = [
+        # List of allowed peers.
+        { 
+          name = "Apex";
+          publicKey = "VwhcN8vSOzdJEotQTpmPHBC52x3Hbv1lkFIyKubrnUA=";
+          # List of IPs assigned to this peer within the tunnel subnet. Used to configure routing.
+          allowedIPs = [ "10.106.0.30/32" ];
+        }
+      ];
+    };
+  };
+
+  networking.hosts = {
+    "10.106.0.30" = [ "apex" ];
+  };
+
+  networking.firewall = {
+    extraCommands = ''
+      # Accept slurm connections to slurmd from apex (via wireguard)
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.30/32 -d 10.106.0.1/32 --dport 6818 -j nixos-fw-accept
+    '';
+  };
+}

m/hut/configuration.nix

@@ -7,11 +7,9 @@
     ../module/ceph.nix
     ../module/debuginfod.nix
     ../module/emulation.nix
-    ../module/slurm-client.nix
     ./gitlab-runner.nix
     ./monitoring.nix
     ./nfs.nix
-    ./slurm-server.nix
     ./nix-serve.nix
     ./public-inbox.nix
     ./gitea.nix

m/hut/slurm-server.nix

@@ -1,7 +0,0 @@
-{ ... }:
-
-{
-  services.slurm = {
-    server.enable = true;
-  };
-}

m/module/nvidia.nix

@@ -0,0 +1,20 @@
+{ lib, config, pkgs, ... }:
+{
+  # Configure Nvidia driver to use with CUDA
+  hardware.nvidia.package = config.boot.kernelPackages.nvidiaPackages.production;
+  hardware.nvidia.open = lib.mkDefault (builtins.abort "hardware.nvidia.open not set");
+  hardware.graphics.enable = true;
+  nixpkgs.config.nvidia.acceptLicense = true;
+  services.xserver.videoDrivers = [ "nvidia" ];
+
+  # enable support for derivations which require nvidia-gpu to be available
+  # > requiredSystemFeatures = [ "cuda" ];
+  programs.nix-required-mounts.enable = true;
+  programs.nix-required-mounts.presets.nvidia-gpu.enable = true;
+  # They forgot to add the symlink
+  programs.nix-required-mounts.allowedPatterns.nvidia-gpu.paths = [
+    config.systemd.tmpfiles.settings.graphics-driver."/run/opengl-driver"."L+".argument
+  ];
+
+  environment.systemPackages = [ pkgs.cudainfo ];
+}

m/module/power-policy.nix

@@ -0,0 +1,33 @@
+{ config, lib, pkgs, ... }:
+
+with lib;
+
+let
+  cfg = config.power.policy;
+in
+{
+  options = {
+    power.policy = mkOption {
+      type = types.nullOr (types.enum [ "always-on" "previous" "always-off" ]);
+      default = null;
+      description = "Set power policy to use via IPMI.";
+    };
+  };
+
+  config = mkIf (cfg != null) {
+    systemd.services."power-policy" = {
+      description = "Set power policy to use via IPMI";
+      wantedBy = [ "multi-user.target" ];
+      unitConfig = {
+        StartLimitBurst = "10";
+        StartLimitIntervalSec = "10m";
+      };
+      serviceConfig = {
+        ExecStart = "${pkgs.ipmitool}/bin/ipmitool chassis policy ${cfg}";
+        Type = "oneshot";
+        Restart = "on-failure";
+        RestartSec = "5s";
+      };
+    };
+  };
+}

m/module/slurm-client.nix

@@ -1,33 +1,10 @@
-{ config, pkgs, lib, ... }:
+{ lib, ... }:
 
-let
+{
-  suspendProgram = pkgs.writeScript "suspend.sh" ''
+  imports = [
-    #!/usr/bin/env bash
+    ./slurm-common.nix
-    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
+  ];
-    set -x
-    export "PATH=/run/current-system/sw/bin:$PATH"
-    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
-    hosts=$(scontrol show hostnames $1)
-    for host in $hosts; do
-      echo Shutting down host: $host
-      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power off
-    done
-  '';
 
-  resumeProgram = pkgs.writeScript "resume.sh" ''
-    #!/usr/bin/env bash
-    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
-    set -x
-    export "PATH=/run/current-system/sw/bin:$PATH"
-    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
-    hosts=$(scontrol show hostnames $1)
-    for host in $hosts; do
-      echo Starting host: $host
-      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power on
-    done
-  '';
-
-in {
   systemd.services.slurmd.serviceConfig = {
     # Kill all processes in the control group on stop/restart. This will kill
     # all the jobs running, so ensure that we only upgrade when the nodes are
@@ -37,90 +14,5 @@ in {
     KillMode = lib.mkForce "control-group";
   };
 
-  services.slurm = {
+  services.slurm.client.enable = true;
-    client.enable = true;
-    controlMachine = "hut";
-    clusterName = "jungle";
-    nodeName = [
-      "owl[1,2]  Sockets=2 CoresPerSocket=14 ThreadsPerCore=2 Feature=owl"
-      "hut       Sockets=2 CoresPerSocket=14 ThreadsPerCore=2"
-    ];
-
-    partitionName = [
-      "owl Nodes=owl[1-2]     Default=YES DefaultTime=01:00:00 MaxTime=INFINITE State=UP"
-    ];
-
-    # See slurm.conf(5) for more details about these options.
-    extraConfig = ''
-      # Use PMIx for MPI by default. It works okay with MPICH and OpenMPI, but
-      # not with Intel MPI. For that use the compatibility shim libpmi.so
-      # setting I_MPI_PMI_LIBRARY=$pmix/lib/libpmi.so while maintaining the PMIx
-      # library in SLURM (--mpi=pmix). See more details here:
-      # https://pm.bsc.es/gitlab/rarias/jungle/-/issues/16
-      MpiDefault=pmix
-
-      # When a node reboots return that node to the slurm queue as soon as it
-      # becomes operative again.
-      ReturnToService=2
-
-      # Track all processes by using a cgroup
-      ProctrackType=proctrack/cgroup
-
-      # Enable task/affinity to allow the jobs to run in a specified subset of
-      # the resources. Use the task/cgroup plugin to enable process containment.
-      TaskPlugin=task/affinity,task/cgroup
-
-      # Power off unused nodes until they are requested
-      SuspendProgram=${suspendProgram}
-      SuspendTimeout=60
-      ResumeProgram=${resumeProgram}
-      ResumeTimeout=300
-      SuspendExcNodes=hut
-
-      # Turn the nodes off after 1 hour of inactivity
-      SuspendTime=3600
-
-      # Reduce port range so we can allow only this range in the firewall
-      SrunPortRange=60000-61000
-
-      # Use cores as consumable resources. In SLURM terms, a core may have
-      # multiple hardware threads (or CPUs).
-      SelectType=select/cons_tres
-
-      # Ignore memory constraints and only use unused cores to share a node with
-      # other jobs.
-      SelectTypeParameters=CR_Core
-
-      # Required for pam_slurm_adopt, see https://slurm.schedmd.com/pam_slurm_adopt.html
-      # This sets up the "extern" step into which ssh-launched processes will be
-      # adopted. Alloc runs the prolog at job allocation (salloc) rather than
-      # when a task runs (srun) so we can ssh early.
-      PrologFlags=Alloc,Contain,X11
-
-      # LaunchParameters=ulimit_pam_adopt will set RLIMIT_RSS in processes
-      # adopted by the external step, similar to tasks running in regular steps
-      # LaunchParameters=ulimit_pam_adopt
-      SlurmdDebug=debug5
-      #DebugFlags=Protocol,Cgroup
-    '';
-
-    extraCgroupConfig = ''
-      CgroupPlugin=cgroup/v2
-      #ConstrainCores=yes
-    '';
-  };
-
-  # Place the slurm config in /etc as this will be required by PAM
-  environment.etc.slurm.source = config.services.slurm.etcSlurm;
-
-  age.secrets.mungeKey = {
-    file = ../../secrets/munge-key.age;
-    owner = "munge";
-    group = "munge";
-  };
-
-  services.munge = {
-    enable = true;
-    password = config.age.secrets.mungeKey.path;
-  };
 }

m/module/slurm-common.nix

@@ -0,0 +1,115 @@
+{ config, pkgs, ... }:
+
+let
+  suspendProgram = pkgs.writeShellScript "suspend.sh" ''
+    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
+    set -x
+    export "PATH=/run/current-system/sw/bin:$PATH"
+    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
+    hosts=$(scontrol show hostnames $1)
+    for host in $hosts; do
+      echo Shutting down host: $host
+      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power off
+    done
+  '';
+
+  resumeProgram = pkgs.writeShellScript "resume.sh" ''
+    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
+    set -x
+    export "PATH=/run/current-system/sw/bin:$PATH"
+    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
+    hosts=$(scontrol show hostnames $1)
+    for host in $hosts; do
+      echo Starting host: $host
+      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power on
+    done
+  '';
+
+in {
+  services.slurm = {
+    controlMachine = "apex";
+    clusterName = "jungle";
+    nodeName = [
+      "owl[1,2]  Sockets=2 CoresPerSocket=14 ThreadsPerCore=2 Feature=owl"
+      "fox       Sockets=8 CoresPerSocket=24 ThreadsPerCore=1"
+    ];
+
+    partitionName = [
+      "owl Nodes=owl[1-2]     Default=YES DefaultTime=01:00:00 MaxTime=INFINITE State=UP"
+      "fox Nodes=fox          Default=NO  DefaultTime=01:00:00 MaxTime=INFINITE State=UP"
+    ];
+
+    # See slurm.conf(5) for more details about these options.
+    extraConfig = ''
+      # Use PMIx for MPI by default. It works okay with MPICH and OpenMPI, but
+      # not with Intel MPI. For that use the compatibility shim libpmi.so
+      # setting I_MPI_PMI_LIBRARY=$pmix/lib/libpmi.so while maintaining the PMIx
+      # library in SLURM (--mpi=pmix). See more details here:
+      # https://pm.bsc.es/gitlab/rarias/jungle/-/issues/16
+      MpiDefault=pmix
+
+      # When a node reboots return that node to the slurm queue as soon as it
+      # becomes operative again.
+      ReturnToService=2
+
+      # Track all processes by using a cgroup
+      ProctrackType=proctrack/cgroup
+
+      # Enable task/affinity to allow the jobs to run in a specified subset of
+      # the resources. Use the task/cgroup plugin to enable process containment.
+      TaskPlugin=task/affinity,task/cgroup
+
+      # Power off unused nodes until they are requested
+      SuspendProgram=${suspendProgram}
+      SuspendTimeout=60
+      ResumeProgram=${resumeProgram}
+      ResumeTimeout=300
+      SuspendExcNodes=fox
+
+      # Turn the nodes off after 1 hour of inactivity
+      SuspendTime=3600
+
+      # Reduce port range so we can allow only this range in the firewall
+      SrunPortRange=60000-61000
+
+      # Use cores as consumable resources. In SLURM terms, a core may have
+      # multiple hardware threads (or CPUs).
+      SelectType=select/cons_tres
+
+      # Ignore memory constraints and only use unused cores to share a node with
+      # other jobs.
+      SelectTypeParameters=CR_Core
+
+      # Required for pam_slurm_adopt, see https://slurm.schedmd.com/pam_slurm_adopt.html
+      # This sets up the "extern" step into which ssh-launched processes will be
+      # adopted. Alloc runs the prolog at job allocation (salloc) rather than
+      # when a task runs (srun) so we can ssh early.
+      PrologFlags=Alloc,Contain,X11
+
+      # LaunchParameters=ulimit_pam_adopt will set RLIMIT_RSS in processes
+      # adopted by the external step, similar to tasks running in regular steps
+      # LaunchParameters=ulimit_pam_adopt
+      SlurmdDebug=debug5
+      #DebugFlags=Protocol,Cgroup
+    '';
+
+    extraCgroupConfig = ''
+      CgroupPlugin=cgroup/v2
+      #ConstrainCores=yes
+    '';
+  };
+
+  # Place the slurm config in /etc as this will be required by PAM
+  environment.etc.slurm.source = config.services.slurm.etcSlurm;
+
+  age.secrets.mungeKey = {
+    file = ../../secrets/munge-key.age;
+    owner = "munge";
+    group = "munge";
+  };
+
+  services.munge = {
+    enable = true;
+    password = config.age.secrets.mungeKey.path;
+  };
+}

m/module/slurm-server.nix

@@ -0,0 +1,23 @@
+{ ... }:
+
+{
+  imports = [
+    ./slurm-common.nix
+  ];
+
+  services.slurm.server.enable = true;
+
+  networking.firewall = {
+    extraCommands = ''
+      # Accept slurm connections to controller from compute nodes
+      iptables -A nixos-fw -p tcp -s 10.0.40.0/24 --dport 6817 -j nixos-fw-accept
+      # Accept slurm connections from compute nodes for srun
+      iptables -A nixos-fw -p tcp -s 10.0.40.0/24 --dport 60000:61000 -j nixos-fw-accept
+
+      # Accept slurm connections to controller from fox (via wireguard)
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.1/32 --dport 6817 -j nixos-fw-accept
+      # Accept slurm connections from fox for srun (via wireguard)
+      iptables -A nixos-fw -p tcp -i wg0 -s 10.106.0.1/32 --dport 60000:61000 -j nixos-fw-accept
+    '';
+  };
+}

m/raccoon/configuration.nix

@@ -6,6 +6,7 @@
     ../module/emulation.nix
     ../module/debuginfod.nix
     ../module/ssh-hut-extern.nix
+    ../module/nvidia.nix
     ../eudy/kernel/perf.nix
   ];
 
@@ -38,6 +39,7 @@
     };
     hosts = {
       "10.0.44.4" = [ "tent" ];
+      "84.88.53.236" = [ "apex" ];
     };
   };
 
@@ -49,12 +51,7 @@
   # Enable performance governor
   powerManagement.cpuFreqGovernor = "performance";
 
-  # Configure Nvidia driver to use with CUDA
-  hardware.nvidia.package = config.boot.kernelPackages.nvidiaPackages.production;
   hardware.nvidia.open = false; # Maxwell is older than Turing architecture
-  hardware.graphics.enable = true;
-  nixpkgs.config.nvidia.acceptLicense = true;
-  services.xserver.videoDrivers = [ "nvidia" ];
 
   services.openssh.settings.X11Forwarding = true;
 

m/tent/configuration.nix

@@ -33,6 +33,9 @@
     nameservers = [ "84.88.52.35" "84.88.52.36" ];
     search = [ "bsc.es" "ac.upc.edu" ];
     defaultGateway = "10.0.44.1";
+    hosts = {
+      "84.88.53.236" = [ "apex" ];
+    };
   };
 
   services.p.enable = true;

m/tent/gitea.nix

@@ -26,5 +26,7 @@
         SENDMAIL_ARGS = "--";
       };
     };
+
+    lfs.enable = true;
   };
 }

m/tent/nginx.nix

@@ -39,6 +39,7 @@ in
           rewrite ^/git/(.*) /$1 break;
           proxy_pass http://127.0.0.1:3000;
           proxy_redirect http:// $scheme://;
+          client_max_body_size 64M;
         }
         location /cache {
           rewrite ^/cache/(.*) /$1 break;

m/weasel/configuration.nix

@@ -14,6 +14,10 @@
   # Users with sudo access
   users.groups.wheel.members = [ "abonerib" "anavarro" ];
 
+  # Run julia installed with juliaup using julia's own libraries:
+  # NIX_LD_LIBRARY_PATH=~/.julia/juliaup/${VERS}/lib/julia ~/.juliaup/bin/julia
+  programs.nix-ld.enable = true;
+
   networking = {
     hostName = "weasel";
     interfaces.eno1.ipv4.addresses = [ {

pkgs/cudainfo/Makefile

@@ -0,0 +1,12 @@
+HOSTCXX  ?= g++
+NVCC     := nvcc -ccbin $(HOSTCXX)
+CXXFLAGS := -m64
+
+# Target rules
+all: cudainfo
+
+cudainfo: cudainfo.cpp
+	$(NVCC) $(CXXFLAGS) -o $@ $<
+
+clean:
+	rm -f cudainfo cudainfo.o

pkgs/cudainfo/cudainfo.cpp

@@ -0,0 +1,600 @@
+/*
+ * Copyright 1993-2015 NVIDIA Corporation.  All rights reserved.
+ *
+ * Please refer to the NVIDIA end user license agreement (EULA) associated
+ * with this source code for terms and conditions that govern your use of
+ * this software. Any use, reproduction, disclosure, or distribution of
+ * this software and related documentation outside the terms of the EULA
+ * is strictly prohibited.
+ *
+ */
+/* This sample queries the properties of the CUDA devices present in the system via CUDA Runtime API. */
+
+// Shared Utilities (QA Testing)
+
+// std::system includes
+#include <memory>
+#include <iostream>
+
+#include <cuda_runtime.h>
+
+// This will output the proper CUDA error strings in the event that a CUDA host call returns an error
+#define checkCudaErrors(val)           check ( (val), #val, __FILE__, __LINE__ )
+
+// CUDA Runtime error messages
+#ifdef __DRIVER_TYPES_H__
+static const char *_cudaGetErrorEnum(cudaError_t error)
+{
+    switch (error)
+    {
+        case cudaSuccess:
+            return "cudaSuccess";
+
+        case cudaErrorMissingConfiguration:
+            return "cudaErrorMissingConfiguration";
+
+        case cudaErrorMemoryAllocation:
+            return "cudaErrorMemoryAllocation";
+
+        case cudaErrorInitializationError:
+            return "cudaErrorInitializationError";
+
+        case cudaErrorLaunchFailure:
+            return "cudaErrorLaunchFailure";
+
+        case cudaErrorPriorLaunchFailure:
+            return "cudaErrorPriorLaunchFailure";
+
+        case cudaErrorLaunchTimeout:
+            return "cudaErrorLaunchTimeout";
+
+        case cudaErrorLaunchOutOfResources:
+            return "cudaErrorLaunchOutOfResources";
+
+        case cudaErrorInvalidDeviceFunction:
+            return "cudaErrorInvalidDeviceFunction";
+
+        case cudaErrorInvalidConfiguration:
+            return "cudaErrorInvalidConfiguration";
+
+        case cudaErrorInvalidDevice:
+            return "cudaErrorInvalidDevice";
+
+        case cudaErrorInvalidValue:
+            return "cudaErrorInvalidValue";
+
+        case cudaErrorInvalidPitchValue:
+            return "cudaErrorInvalidPitchValue";
+
+        case cudaErrorInvalidSymbol:
+            return "cudaErrorInvalidSymbol";
+
+        case cudaErrorMapBufferObjectFailed:
+            return "cudaErrorMapBufferObjectFailed";
+
+        case cudaErrorUnmapBufferObjectFailed:
+            return "cudaErrorUnmapBufferObjectFailed";
+
+        case cudaErrorInvalidHostPointer:
+            return "cudaErrorInvalidHostPointer";
+
+        case cudaErrorInvalidDevicePointer:
+            return "cudaErrorInvalidDevicePointer";
+
+        case cudaErrorInvalidTexture:
+            return "cudaErrorInvalidTexture";
+
+        case cudaErrorInvalidTextureBinding:
+            return "cudaErrorInvalidTextureBinding";
+
+        case cudaErrorInvalidChannelDescriptor:
+            return "cudaErrorInvalidChannelDescriptor";
+
+        case cudaErrorInvalidMemcpyDirection:
+            return "cudaErrorInvalidMemcpyDirection";
+
+        case cudaErrorAddressOfConstant:
+            return "cudaErrorAddressOfConstant";
+
+        case cudaErrorTextureFetchFailed:
+            return "cudaErrorTextureFetchFailed";
+
+        case cudaErrorTextureNotBound:
+            return "cudaErrorTextureNotBound";
+
+        case cudaErrorSynchronizationError:
+            return "cudaErrorSynchronizationError";
+
+        case cudaErrorInvalidFilterSetting:
+            return "cudaErrorInvalidFilterSetting";
+
+        case cudaErrorInvalidNormSetting:
+            return "cudaErrorInvalidNormSetting";
+
+        case cudaErrorMixedDeviceExecution:
+            return "cudaErrorMixedDeviceExecution";
+
+        case cudaErrorCudartUnloading:
+            return "cudaErrorCudartUnloading";
+
+        case cudaErrorUnknown:
+            return "cudaErrorUnknown";
+
+        case cudaErrorNotYetImplemented:
+            return "cudaErrorNotYetImplemented";
+
+        case cudaErrorMemoryValueTooLarge:
+            return "cudaErrorMemoryValueTooLarge";
+
+        case cudaErrorInvalidResourceHandle:
+            return "cudaErrorInvalidResourceHandle";
+
+        case cudaErrorNotReady:
+            return "cudaErrorNotReady";
+
+        case cudaErrorInsufficientDriver:
+            return "cudaErrorInsufficientDriver";
+
+        case cudaErrorSetOnActiveProcess:
+            return "cudaErrorSetOnActiveProcess";
+
+        case cudaErrorInvalidSurface:
+            return "cudaErrorInvalidSurface";
+
+        case cudaErrorNoDevice:
+            return "cudaErrorNoDevice";
+
+        case cudaErrorECCUncorrectable:
+            return "cudaErrorECCUncorrectable";
+
+        case cudaErrorSharedObjectSymbolNotFound:
+            return "cudaErrorSharedObjectSymbolNotFound";
+
+        case cudaErrorSharedObjectInitFailed:
+            return "cudaErrorSharedObjectInitFailed";
+
+        case cudaErrorUnsupportedLimit:
+            return "cudaErrorUnsupportedLimit";
+
+        case cudaErrorDuplicateVariableName:
+            return "cudaErrorDuplicateVariableName";
+
+        case cudaErrorDuplicateTextureName:
+            return "cudaErrorDuplicateTextureName";
+
+        case cudaErrorDuplicateSurfaceName:
+            return "cudaErrorDuplicateSurfaceName";
+
+        case cudaErrorDevicesUnavailable:
+            return "cudaErrorDevicesUnavailable";
+
+        case cudaErrorInvalidKernelImage:
+            return "cudaErrorInvalidKernelImage";
+
+        case cudaErrorNoKernelImageForDevice:
+            return "cudaErrorNoKernelImageForDevice";
+
+        case cudaErrorIncompatibleDriverContext:
+            return "cudaErrorIncompatibleDriverContext";
+
+        case cudaErrorPeerAccessAlreadyEnabled:
+            return "cudaErrorPeerAccessAlreadyEnabled";
+
+        case cudaErrorPeerAccessNotEnabled:
+            return "cudaErrorPeerAccessNotEnabled";
+
+        case cudaErrorDeviceAlreadyInUse:
+            return "cudaErrorDeviceAlreadyInUse";
+
+        case cudaErrorProfilerDisabled:
+            return "cudaErrorProfilerDisabled";
+
+        case cudaErrorProfilerNotInitialized:
+            return "cudaErrorProfilerNotInitialized";
+
+        case cudaErrorProfilerAlreadyStarted:
+            return "cudaErrorProfilerAlreadyStarted";
+
+        case cudaErrorProfilerAlreadyStopped:
+            return "cudaErrorProfilerAlreadyStopped";
+
+        /* Since CUDA 4.0*/
+        case cudaErrorAssert:
+            return "cudaErrorAssert";
+
+        case cudaErrorTooManyPeers:
+            return "cudaErrorTooManyPeers";
+
+        case cudaErrorHostMemoryAlreadyRegistered:
+            return "cudaErrorHostMemoryAlreadyRegistered";
+
+        case cudaErrorHostMemoryNotRegistered:
+            return "cudaErrorHostMemoryNotRegistered";
+
+        /* Since CUDA 5.0 */
+        case cudaErrorOperatingSystem:
+            return "cudaErrorOperatingSystem";
+
+        case cudaErrorPeerAccessUnsupported:
+            return "cudaErrorPeerAccessUnsupported";
+
+        case cudaErrorLaunchMaxDepthExceeded:
+            return "cudaErrorLaunchMaxDepthExceeded";
+
+        case cudaErrorLaunchFileScopedTex:
+            return "cudaErrorLaunchFileScopedTex";
+
+        case cudaErrorLaunchFileScopedSurf:
+            return "cudaErrorLaunchFileScopedSurf";
+
+        case cudaErrorSyncDepthExceeded:
+            return "cudaErrorSyncDepthExceeded";
+
+        case cudaErrorLaunchPendingCountExceeded:
+            return "cudaErrorLaunchPendingCountExceeded";
+
+        case cudaErrorNotPermitted:
+            return "cudaErrorNotPermitted";
+
+        case cudaErrorNotSupported:
+            return "cudaErrorNotSupported";
+
+        /* Since CUDA 6.0 */
+        case cudaErrorHardwareStackError:
+            return "cudaErrorHardwareStackError";
+
+        case cudaErrorIllegalInstruction:
+            return "cudaErrorIllegalInstruction";
+
+        case cudaErrorMisalignedAddress:
+            return "cudaErrorMisalignedAddress";
+
+        case cudaErrorInvalidAddressSpace:
+            return "cudaErrorInvalidAddressSpace";
+
+        case cudaErrorInvalidPc:
+            return "cudaErrorInvalidPc";
+
+        case cudaErrorIllegalAddress:
+            return "cudaErrorIllegalAddress";
+
+        /* Since CUDA 6.5*/
+        case cudaErrorInvalidPtx:
+            return "cudaErrorInvalidPtx";
+
+        case cudaErrorInvalidGraphicsContext:
+            return "cudaErrorInvalidGraphicsContext";
+
+        case cudaErrorStartupFailure:
+            return "cudaErrorStartupFailure";
+
+        case cudaErrorApiFailureBase:
+            return "cudaErrorApiFailureBase";
+    }
+
+    return "<unknown>";
+}
+#endif
+
+template< typename T >
+void check(T result, char const *const func, const char *const file, int const line)
+{
+    if (result)
+    {
+        fprintf(stderr, "CUDA error at %s:%d code=%d(%s) \"%s\" \n",
+                file, line, static_cast<unsigned int>(result), _cudaGetErrorEnum(result), func);
+        cudaDeviceReset();
+        // Make sure we call CUDA Device Reset before exiting
+        exit(EXIT_FAILURE);
+    }
+}
+
+int *pArgc = NULL;
+char **pArgv = NULL;
+
+#if CUDART_VERSION < 5000
+
+// CUDA-C includes
+#include <cuda.h>
+
+// This function wraps the CUDA Driver API into a template function
+template <class T>
+inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute, int device)
+{
+    CUresult error =    cuDeviceGetAttribute(attribute, device_attribute, device);
+
+    if (CUDA_SUCCESS != error) {
+        fprintf(stderr, "cuSafeCallNoSync() Driver API error = %04d from file <%s>, line %i.\n",
+                error, __FILE__, __LINE__);
+
+        // cudaDeviceReset causes the driver to clean up all state. While
+        // not mandatory in normal operation, it is good practice.  It is also
+        // needed to ensure correct operation when the application is being
+        // profiled. Calling cudaDeviceReset causes all profile data to be
+        // flushed before the application exits
+        cudaDeviceReset();
+        exit(EXIT_FAILURE);
+    }
+}
+
+#endif /* CUDART_VERSION < 5000 */
+
+// Beginning of GPU Architecture definitions
+inline int ConvertSMVer2Cores(int major, int minor)
+{
+    // Defines for GPU Architecture types (using the SM version to determine the # of cores per SM
+    typedef struct {
+        int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM minor version
+        int Cores;
+    } sSMtoCores;
+
+    sSMtoCores nGpuArchCoresPerSM[] = {
+        { 0x20, 32 }, // Fermi Generation (SM 2.0) GF100 class
+        { 0x21, 48 }, // Fermi Generation (SM 2.1) GF10x class
+        { 0x30, 192}, // Kepler Generation (SM 3.0) GK10x class
+        { 0x32, 192}, // Kepler Generation (SM 3.2) GK10x class
+        { 0x35, 192}, // Kepler Generation (SM 3.5) GK11x class
+        { 0x37, 192}, // Kepler Generation (SM 3.7) GK21x class
+        { 0x50, 128}, // Maxwell Generation (SM 5.0) GM10x class
+        { 0x52, 128}, // Maxwell Generation (SM 5.2) GM20x class
+        {   -1, -1 }
+    };
+
+    int index = 0;
+
+    while (nGpuArchCoresPerSM[index].SM != -1) {
+        if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor)) {
+            return nGpuArchCoresPerSM[index].Cores;
+        }
+
+        index++;
+    }
+
+    // If we don't find the values, we default use the previous one to run properly
+    printf("MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n", major, minor, nGpuArchCoresPerSM[index-1].Cores);
+    return nGpuArchCoresPerSM[index-1].Cores;
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Program main
+////////////////////////////////////////////////////////////////////////////////
+int
+main(int argc, char **argv)
+{
+    pArgc = &argc;
+    pArgv = argv;
+
+    printf("%s Starting...\n\n", argv[0]);
+    printf(" CUDA Device Query (Runtime API) version (CUDART static linking)\n\n");
+
+    int deviceCount = 0;
+    cudaError_t error_id = cudaGetDeviceCount(&deviceCount);
+
+    if (error_id != cudaSuccess) {
+        printf("cudaGetDeviceCount failed: %s (%d)\n",
+			cudaGetErrorString(error_id), (int) error_id);
+        printf("Result = FAIL\n");
+        exit(EXIT_FAILURE);
+    }
+
+    // This function call returns 0 if there are no CUDA capable devices.
+    if (deviceCount == 0)
+        printf("There are no available device(s) that support CUDA\n");
+    else
+        printf("Detected %d CUDA Capable device(s)\n", deviceCount);
+
+    int dev, driverVersion = 0, runtimeVersion = 0;
+
+    for (dev = 0; dev < deviceCount; ++dev) {
+        cudaSetDevice(dev);
+        cudaDeviceProp deviceProp;
+        cudaGetDeviceProperties(&deviceProp, dev);
+
+        printf("\nDevice %d: \"%s\"\n", dev, deviceProp.name);
+
+        // Console log
+        cudaDriverGetVersion(&driverVersion);
+        cudaRuntimeGetVersion(&runtimeVersion);
+        printf("  CUDA Driver Version / Runtime Version          %d.%d / %d.%d\n", driverVersion/1000, (driverVersion%100)/10, runtimeVersion/1000, (runtimeVersion%100)/10);
+        printf("  CUDA Capability Major/Minor version number:    %d.%d\n", deviceProp.major, deviceProp.minor);
+
+        printf("  Total amount of global memory:                 %.0f MBytes (%llu bytes)\n",
+                (float)deviceProp.totalGlobalMem/1048576.0f, (unsigned long long) deviceProp.totalGlobalMem);
+
+        printf("  (%2d) Multiprocessors, (%3d) CUDA Cores/MP:     %d CUDA Cores\n",
+               deviceProp.multiProcessorCount,
+               ConvertSMVer2Cores(deviceProp.major, deviceProp.minor),
+               ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) * deviceProp.multiProcessorCount);
+        printf("  GPU Max Clock rate:                            %.0f MHz (%0.2f GHz)\n", deviceProp.clockRate * 1e-3f, deviceProp.clockRate * 1e-6f);
+
+
+#if CUDART_VERSION >= 5000
+        // This is supported in CUDA 5.0 (runtime API device properties)
+        printf("  Memory Clock rate:                             %.0f Mhz\n", deviceProp.memoryClockRate * 1e-3f);
+        printf("  Memory Bus Width:                              %d-bit\n",   deviceProp.memoryBusWidth);
+
+        if (deviceProp.l2CacheSize) {
+            printf("  L2 Cache Size:                                 %d bytes\n", deviceProp.l2CacheSize);
+        }
+
+#else
+        // This only available in CUDA 4.0-4.2 (but these were only exposed in the CUDA Driver API)
+        int memoryClock;
+        getCudaAttribute<int>(&memoryClock, CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE, dev);
+        printf("  Memory Clock rate:                             %.0f Mhz\n", memoryClock * 1e-3f);
+        int memBusWidth;
+        getCudaAttribute<int>(&memBusWidth, CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH, dev);
+        printf("  Memory Bus Width:                              %d-bit\n", memBusWidth);
+        int L2CacheSize;
+        getCudaAttribute<int>(&L2CacheSize, CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE, dev);
+
+        if (L2CacheSize) {
+            printf("  L2 Cache Size:                                 %d bytes\n", L2CacheSize);
+        }
+
+#endif
+
+        printf("  Maximum Texture Dimension Size (x,y,z)         1D=(%d), 2D=(%d, %d), 3D=(%d, %d, %d)\n",
+               deviceProp.maxTexture1D   , deviceProp.maxTexture2D[0], deviceProp.maxTexture2D[1],
+               deviceProp.maxTexture3D[0], deviceProp.maxTexture3D[1], deviceProp.maxTexture3D[2]);
+        printf("  Maximum Layered 1D Texture Size, (num) layers  1D=(%d), %d layers\n",
+               deviceProp.maxTexture1DLayered[0], deviceProp.maxTexture1DLayered[1]);
+        printf("  Maximum Layered 2D Texture Size, (num) layers  2D=(%d, %d), %d layers\n",
+               deviceProp.maxTexture2DLayered[0], deviceProp.maxTexture2DLayered[1], deviceProp.maxTexture2DLayered[2]);
+
+
+        printf("  Total amount of constant memory:               %lu bytes\n", deviceProp.totalConstMem);
+        printf("  Total amount of shared memory per block:       %lu bytes\n", deviceProp.sharedMemPerBlock);
+        printf("  Total number of registers available per block: %d\n", deviceProp.regsPerBlock);
+        printf("  Warp size:                                     %d\n", deviceProp.warpSize);
+        printf("  Maximum number of threads per multiprocessor:  %d\n", deviceProp.maxThreadsPerMultiProcessor);
+        printf("  Maximum number of threads per block:           %d\n", deviceProp.maxThreadsPerBlock);
+        printf("  Max dimension size of a thread block (x,y,z): (%d, %d, %d)\n",
+               deviceProp.maxThreadsDim[0],
+               deviceProp.maxThreadsDim[1],
+               deviceProp.maxThreadsDim[2]);
+        printf("  Max dimension size of a grid size    (x,y,z): (%d, %d, %d)\n",
+               deviceProp.maxGridSize[0],
+               deviceProp.maxGridSize[1],
+               deviceProp.maxGridSize[2]);
+        printf("  Maximum memory pitch:                          %lu bytes\n", deviceProp.memPitch);
+        printf("  Texture alignment:                             %lu bytes\n", deviceProp.textureAlignment);
+        printf("  Concurrent copy and kernel execution:          %s with %d copy engine(s)\n", (deviceProp.deviceOverlap ? "Yes" : "No"), deviceProp.asyncEngineCount);
+        printf("  Run time limit on kernels:                     %s\n", deviceProp.kernelExecTimeoutEnabled ? "Yes" : "No");
+        printf("  Integrated GPU sharing Host Memory:            %s\n", deviceProp.integrated ? "Yes" : "No");
+        printf("  Support host page-locked memory mapping:       %s\n", deviceProp.canMapHostMemory ? "Yes" : "No");
+        printf("  Alignment requirement for Surfaces:            %s\n", deviceProp.surfaceAlignment ? "Yes" : "No");
+        printf("  Device has ECC support:                        %s\n", deviceProp.ECCEnabled ? "Enabled" : "Disabled");
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+        printf("  CUDA Device Driver Mode (TCC or WDDM):         %s\n", deviceProp.tccDriver ? "TCC (Tesla Compute Cluster Driver)" : "WDDM (Windows Display Driver Model)");
+#endif
+        printf("  Device supports Unified Addressing (UVA):      %s\n", deviceProp.unifiedAddressing ? "Yes" : "No");
+        printf("  Device PCI Domain ID / Bus ID / location ID:   %d / %d / %d\n", deviceProp.pciDomainID, deviceProp.pciBusID, deviceProp.pciDeviceID);
+
+        const char *sComputeMode[] = {
+            "Default (multiple host threads can use ::cudaSetDevice() with device simultaneously)",
+            "Exclusive (only one host thread in one process is able to use ::cudaSetDevice() with this device)",
+            "Prohibited (no host thread can use ::cudaSetDevice() with this device)",
+            "Exclusive Process (many threads in one process is able to use ::cudaSetDevice() with this device)",
+            "Unknown",
+            NULL
+        };
+        printf("  Compute Mode:\n");
+        printf("     < %s >\n", sComputeMode[deviceProp.computeMode]);
+    }
+
+    // If there are 2 or more GPUs, query to determine whether RDMA is supported
+    if (deviceCount >= 2)
+    {
+        cudaDeviceProp prop[64];
+        int gpuid[64]; // we want to find the first two GPU's that can support P2P
+        int gpu_p2p_count = 0;
+
+        for (int i=0; i < deviceCount; i++)
+        {
+            checkCudaErrors(cudaGetDeviceProperties(&prop[i], i));
+
+            // Only boards based on Fermi or later can support P2P
+            if ((prop[i].major >= 2)
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+                // on Windows (64-bit), the Tesla Compute Cluster driver for windows must be enabled to supprot this
+                && prop[i].tccDriver
+#endif
+               )
+            {
+                // This is an array of P2P capable GPUs
+                gpuid[gpu_p2p_count++] = i;
+            }
+        }
+
+        // Show all the combinations of support P2P GPUs
+        int can_access_peer_0_1, can_access_peer_1_0;
+
+        if (gpu_p2p_count >= 2)
+        {
+            for (int i = 0; i < gpu_p2p_count-1; i++)
+            {
+                for (int j = 1; j < gpu_p2p_count; j++)
+                {
+                    checkCudaErrors(cudaDeviceCanAccessPeer(&can_access_peer_0_1, gpuid[i], gpuid[j]));
+                    printf("> Peer access from %s (GPU%d) -> %s (GPU%d) : %s\n", prop[gpuid[i]].name, gpuid[i],
+                           prop[gpuid[j]].name, gpuid[j] ,
+                           can_access_peer_0_1 ? "Yes" : "No");
+                }
+            }
+
+            for (int j = 1; j < gpu_p2p_count; j++)
+            {
+                for (int i = 0; i < gpu_p2p_count-1; i++)
+                {
+                    checkCudaErrors(cudaDeviceCanAccessPeer(&can_access_peer_1_0, gpuid[j], gpuid[i]));
+                    printf("> Peer access from %s (GPU%d) -> %s (GPU%d) : %s\n", prop[gpuid[j]].name, gpuid[j],
+                           prop[gpuid[i]].name, gpuid[i] ,
+                           can_access_peer_1_0 ? "Yes" : "No");
+                }
+            }
+        }
+    }
+
+    // csv masterlog info
+    // *****************************
+    // exe and CUDA driver name
+    printf("\n");
+    std::string sProfileString = "deviceQuery, CUDA Driver = CUDART";
+    char cTemp[128];
+
+    // driver version
+    sProfileString += ", CUDA Driver Version = ";
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+    sprintf_s(cTemp, 10, "%d.%d", driverVersion/1000, (driverVersion%100)/10);
+#else
+    sprintf(cTemp, "%d.%d", driverVersion/1000, (driverVersion%100)/10);
+#endif
+    sProfileString +=  cTemp;
+
+    // Runtime version
+    sProfileString += ", CUDA Runtime Version = ";
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+    sprintf_s(cTemp, 10, "%d.%d", runtimeVersion/1000, (runtimeVersion%100)/10);
+#else
+    sprintf(cTemp, "%d.%d", runtimeVersion/1000, (runtimeVersion%100)/10);
+#endif
+    sProfileString +=  cTemp;
+
+    // Device count
+    sProfileString += ", NumDevs = ";
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+    sprintf_s(cTemp, 10, "%d", deviceCount);
+#else
+    sprintf(cTemp, "%d", deviceCount);
+#endif
+    sProfileString += cTemp;
+
+    // Print Out all device Names
+    for (dev = 0; dev < deviceCount; ++dev)
+    {
+#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
+        sprintf_s(cTemp, 13, ", Device%d = ", dev);
+#else
+        sprintf(cTemp, ", Device%d = ", dev);
+#endif
+        cudaDeviceProp deviceProp;
+        cudaGetDeviceProperties(&deviceProp, dev);
+        sProfileString += cTemp;
+        sProfileString += deviceProp.name;
+    }
+
+    sProfileString += "\n";
+    printf("%s", sProfileString.c_str());
+
+    printf("Result = PASS\n");
+
+    // finish
+    // cudaDeviceReset causes the driver to clean up all state. While
+    // not mandatory in normal operation, it is good practice.  It is also
+    // needed to ensure correct operation when the application is being
+    // profiled. Calling cudaDeviceReset causes all profile data to be
+    // flushed before the application exits
+    cudaDeviceReset();
+    return 0;
+}

pkgs/cudainfo/default.nix

@@ -0,0 +1,43 @@
+{
+  stdenv
+, cudatoolkit
+, cudaPackages
+, autoAddDriverRunpath
+, strace
+}:
+
+stdenv.mkDerivation (finalAttrs: {
+  name = "cudainfo";
+  src = ./.;
+  buildInputs = [
+    cudatoolkit # Required for nvcc
+    cudaPackages.cuda_cudart.static # Required for -lcudart_static
+    autoAddDriverRunpath
+  ];
+  installPhase = ''
+    mkdir -p $out/bin
+    cp -a cudainfo $out/bin
+  '';
+  passthru.gpuCheck = stdenv.mkDerivation {
+    name = "cudainfo-test";
+    requiredSystemFeatures = [ "cuda" ];
+    dontBuild = true;
+    nativeCheckInputs = [
+      finalAttrs.finalPackage # The cudainfo package from above
+      strace # When it fails, it will show the trace
+    ];
+    dontUnpack = true;
+    doCheck = true;
+    checkPhase = ''
+      if ! cudainfo; then
+        set -x
+        cudainfo=$(command -v cudainfo)
+        ldd $cudainfo
+        readelf -d $cudainfo
+        strace -f $cudainfo
+        set +x
+      fi
+    '';
+    installPhase = "touch $out";
+  };
+})

pkgs/overlay.nix

@@ -52,4 +52,5 @@ final: prev:
   prometheus-slurm-exporter = prev.callPackage ./slurm-exporter.nix { };
   meteocat-exporter = prev.callPackage ./meteocat-exporter/default.nix { };
   upc-qaire-exporter = prev.callPackage ./upc-qaire-exporter/default.nix { };
+  cudainfo = prev.callPackage ./cudainfo/default.nix { };
 }

secrets/gitea-runner-token.age

@@ -1,11 +1,13 @@
 age-encryption.org/v1
--> ssh-ed25519 HY2yRg d7+nvfAcdC3GjJxipXFrsfGGyP5jAY+gRWRV+4FVYAM
+-> ssh-ed25519 HY2yRg gKGxsjHfpiRDQ6Tuvcx7pjKgrVUGweotuplLYwCGvik
-CG7r0bRGgnUWcdfDnpe7HwZ3L/y7b5iuJuqvf15b3/Y
+DSz9j/stVyB1lXpVP+kg+H+RDgSftREGFFLQZClC3kI
--> ssh-ed25519 CAWG4Q X0vITOErz4wkR3VQYOcVlnrkHtwe+ytdZz1Hcrs4vVs
+-> ssh-ed25519 cK5kHw 17DpKekfNVy4V742QSd61r2w6iawtOJR7Ct3UflDXio
-6IWYOhXLQ+BnML9YfLLHJYEO2CZ/uEc9IBqhoWvjDHI
+hsqTEPCYjHKvndMWPl4GpG23CzjGgVrS+cLIymISJHU
--> ssh-ed25519 xA739A p5e/0AJtZ0+zbRvkB/usLuxusY8xXRx9Ksi/LQlcIHw
+-> ssh-ed25519 CAWG4Q oK01d4pbBqEZVsymSiKijPvJo714xsMSRMbzkssJKiw
-M4S/qlzT9POyJx4gY9lmycstUcdwG2cinN4OlV22zzo
+hs0tVFkqtIHXg9jtC2iDgCtefFcWvGJkXB+HJUcqXQs
--> ssh-ed25519 MSF3dg Ydl7uBWzBx6sAaxbzC3x8qiaU3ysGqV4rUFLpHCEV30
+-> ssh-ed25519 xA739A KxO+AawfLMERHwzt3YnZRwPFlCfGETma7fo8M+ZtsAY
-/1AUHBhCNOs9i7LJbmzwQDHsu+ybzYf6+coztKk5E3U
+eSn0+/rhLQxNKt5xKubKck8Nxun2Sh3eJqBU/hwgzZM
---- kYt15WxClpT7PXD1oFe9GqJU+OswjH7y9wIc8/GzZ7M
+-> ssh-ed25519 MSF3dg OyaZBLB2kO8fU139lXbbC404gT7IzIWk+BMhYzabBDg
-<EFBFBD><EFBFBD>h<>ߓ<><DF93><EFBFBD>`<60><><EFBFBD>V4F<34><46>_k)^<5E>m$uj:ѳ<><D1B3><17><><EFBFBD>}<7D>Z]$U]<12>u<EFBFBD> <20>0<EFBFBD><30><EFBFBD>v8<76>?<3F>X<EFBFBD>P<EFBFBD>g%d<>#<23>d9{rAi<41><69>
+/fiPFfBJcb+e40+fZbwCw7niF2hh+JxUPiKSiwUSOWg
+--- ycZyGX+Li+LsOuweF9OVPl8aoMaRgp/RdFbDrPszkUs
+<EFBFBD><EFBFBD><EFBFBD><EFBFBD>YM<EFBFBD><EFBFBD>:E O<><4F>2<EFBFBD>r=<15>&4<><04>CQΣ<51><CEA3>hC<68><43><EFBFBD>cb<63>^Sy<53><79>%	<09><>x-vC`g<><15><><EFBFBD><EFBFBD>W^<5E><>wVG<0B><><EFBFBD>

secrets/secrets.nix

@@ -2,6 +2,8 @@ let
   keys = import ../keys.nix;
   adminsKeys = builtins.attrValues keys.admins;
   hut = [ keys.hosts.hut ] ++ adminsKeys;
+  fox = [ keys.hosts.fox ] ++ adminsKeys;
+  apex = [ keys.hosts.apex ] ++ adminsKeys;
   mon = [ keys.hosts.hut keys.hosts.tent ] ++ adminsKeys;
   tent = [ keys.hosts.tent ] ++ adminsKeys;
   # Only expose ceph keys to safe nodes and admins
@@ -24,4 +26,7 @@ in
 
   "ceph-user.age".publicKeys = safe;
   "munge-key.age".publicKeys = safe;
+
+  "wg-fox.age".publicKeys = fox;
+  "wg-apex.age".publicKeys = apex;
 }

secrets/tent-gitlab-runner-bsc-docker-token.age

@@ -1,11 +1,13 @@
 age-encryption.org/v1
--> ssh-ed25519 G5LX5w HlQ4V8lBd3im5j8KHEuQZBTuztvPj1QoWdv6FL6qzGI
+-> ssh-ed25519 G5LX5w Zhbs+NM/SI49qQ0X8bBpWUWxYM0vUKCXNAnPpIE2NR0
-Jpt91X1UIIVFQt1X6Q//kALn+Cetp/LqBZZvTuhFthw
+CkBUmJ26EkwHztT8Pz0UGq2KZwN0Xz8iYQ9cEHL9OWQ
--> ssh-ed25519 CAWG4Q StnngJAcuAwUnTrXDR3nJ2KFN0jNdTqSz+/1TfmWkzA
+-> ssh-ed25519 cK5kHw 5KjUXJywRDp2A7l5ukTCS+WIAalxwP1f71ejGxwNrX4
-CR4AQ6fqaJVY1mdUIX1gzaZwRs1sU8F8hHztnkN8vN0
+JW8OLmfkULXo9AwYMGNyOgZ+nQ0MVc0PCM4kKPIo6V4
--> ssh-ed25519 xA739A xya5A5t63Owx+VrGgUfV/lIP8b/xV1cerMpuZBLaDVM
+-> ssh-ed25519 CAWG4Q cVjY3R0ZHAfokA4kWlu5vOl2Gs7mdqRgRk4WSUOXAjg
-w+pA583yUnFq2AvGBGzWbQIGQEY9WqW0CSLQ9v+SG0c
+IxEDvuximW99EqxmpW+Btpm0Zydmwg/u87bqnl26NYc
--> ssh-ed25519 MSF3dg aXkLxCyYdOwVopHHmpXEI6WlAIizKdJi4IO0KEdhS3s
+-> ssh-ed25519 xA739A hmuwZuxmJnuAjmU4X8yhPQ+hPWvN1G+ZS0pvD7fHamg
-WKXkTszZN66+QZdSDJ4D9q7xgYWMfliOLCubIF2Dqkc
+fnAPW6ZCrv5pSO4RQhhr8xz7ij7jAZJk0ApWluOXDng
---- uVWoU2lMkqQ/9Z0BqKRCeUpsKi8lwmHukT/FV8wYMbg
+-> ssh-ed25519 MSF3dg SSGLcWnum0Qo/0OnKDZVg9xAZMwGwVNYYmRJXxb4GU0
-<EFBFBD><EFBFBD>1G+<2B>6<EFBFBD><36>g[|x]2T<32>й<EFBFBD><D0B9><EFBFBD> <20>CKu)<29><><EFBFBD>]<5D><>8֓<38><D693><EFBFBD><EFBFBD>l<EFBFBD><6C>S<EFBFBD><53><EFBFBD>Q<EFBFBD><07><>x<EFBFBD><78><EFBFBD><EFBFBD>#7r<37>k{*<2A><>3ս~C<>b<EFBFBD><62><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ڵ<EFBFBD>Np<1E><05>]J]h<>je+d%Е<>#<23>m<EFBFBD>?=6}<7D> 
+pdl6kATG7n2oMsoUboBfu+vDKurJcH1UvUa70rfMQkE
+--- a2ZQAeAQlO9DWnegIAq6NpI1Po6f38l+hitZvq+zIW8
+<EFBFBD>\ֺ"^<5E>DT<44>H<EFBFBD><48>3<EFBFBD><33><EFBFBD>_|.h<0E><><EFBFBD><EFBFBD><03>^<5E>n<14><0E><><EFBFBD><EFBFBD><1A>g<EFBFBD>S<EFBFBD>]_<><5F>?n<>z~2<>!<21>p7<70><37><<3C><14>ʨD?<3F>~<02>F<EFBFBD>$<24>`<60>q+<2B><><EFBFBD>SW<53>(+<2B><>P<EFBFBD>c<1E>u[<5B>m<EFBFBD>`O<>ܛ<EFBFBD>ϖT

secrets/tent-gitlab-runner-pm-docker-token.age

@@ -1,11 +1,13 @@
 age-encryption.org/v1
--> ssh-ed25519 G5LX5w sg9SmahxBg35MDIxhrp4oHkaTaxsKoVQju2eNhCt0BM
+-> ssh-ed25519 G5LX5w VKM/Y6Wy0gmb2gc4Q00VzHQ4IAxfSyshuDoaAzlEkFM
-CZ64dEGqz2tbkG8KtimZvLUEMrQpVVBJP7Fu46WTMgc
+vf18uoEN5ZLJ4HcJg85epaseh1CRL9/ncXtU2HpH+QE
--> ssh-ed25519 CAWG4Q jzS1R14W1CWxdziMLG/yCGPLWSkiyE+9lqyCVe491ng
+-> ssh-ed25519 cK5kHw sMuG07kjlI6VjPjELOUPzkn+KT9Yq7BPf0zSATM2aGI
-acJo/nhKq3pSPoFEPaFLN1fzHHbEzstNoLtohWAHKiM
+/eODwL8KwyVgFjBK2MJlbqjN7mEvXCSsjq9D96szrng
--> ssh-ed25519 xA739A qeGJoLeSIQwLU2Yg+Gi2bikHJ3HscLfyo1msqL3JwHw
+-> ssh-ed25519 CAWG4Q t3/Ty7yCqC5x8KQY4VaHSQ9Q3epqMpXoBDKyKx9+VzE
-tTwaxRBKTl/SoyY/LnxR/j/5WvCNX5VeZLKi018YMrY
+JwgUsqMd+1jFZvFp9/SIoowbhSMVEkKp03T69+OHjho
--> ssh-ed25519 MSF3dg Wym7Uyf1XvH1H6mNDERkO8opkMiN0zzXm2PjXftEOWs
+-> ssh-ed25519 xA739A 0ohmKK427+4vupivrtjXp0dDK8wT4XUA9rWgcsCGKgA
-Uw8ZwwKIB5UqgVuoSLE2QajNDJZkH7/Y3Nsy+WFl7Xs
+msbeQyz3pL8RLtAeXX5tsfyHyOXxhfYpqaLEKnRxpPQ
---- 94hGVbYiCGZdMEJesCMLh7IZi+w5l/Kr1lZJHQgrc0o
+-> ssh-ed25519 MSF3dg H+6jAoP7/Dxp8C/7Bk1C4CT1hpkUhtbnTWWIxkO24Ec
-j5j磛<6A><04><>J<EFBFBD><4A><EFBFBD>a<EFBFBD>]<5D>a%dr<64><72>FDT<44><54>^<5E><>Q<EFBFBD>s/<2F>kwB<77>$<24><>$<24><>H<EFBFBD>'<27><><EFBFBD><EFBFBD><EFBFBD>w<14><?^|<7C><07>h$<24>ؗ<EFBFBD>GI<47>ĕsT2RU<52><55>*/O<>7<EFBFBD><37><EFBFBD>G<EFBFBD>pͪ<70>4<EFBFBD><34><EFBFBD>M9<4D>j<><06>
+SrMuUG93T5lUw3xINEen5EEKLXJizIGFhBO1fVroFHE
+--- tIPnH9cxTV3m3qzvZB97Egz+raWwZJ182BXXKDu8f+o
+<EFBFBD><EFBFBD>f#<23>,|<7C>Ey.v<>DL<44>Ӻ<05>J
PX<50><07><>`<60><><EFBFBD><EFBFBD>-#<23>F<EFBFBD>Ubs<62>(Q!?<3F><1A>#xJG?5<><35><EFBFBD><EFBFBD><EFBFBD>~<7E><>6MA<15>U<><55><EFBFBD>C
<01><>M<>$+}W<>NϨG!<21><><EFBFBD><EFBFBD>a<EFBFBD><61><EFBFBD><EFBFBD>%<25>ǽ<EFBFBD>G

secrets/tent-gitlab-runner-pm-shell-token.age

@@ -1,12 +1,13 @@
 age-encryption.org/v1
--> ssh-ed25519 G5LX5w 5K0mzfJGvAB2LGmoQ9ZLbWooVEX6F4+fQdo1JUoB3FM
+-> ssh-ed25519 G5LX5w 1KfTmTRP3iSdcclf/FuIpFWpy1tgKs5ED+qSYWo7inY
-AKGa507bUrYjXFaMQ1MXTDBFYsdS6zbs+flmxYN0UNo
+RX6Q1nLFF/yiVLpkWrl0BI0PpLoBi753+y8l/AXjNE4
--> ssh-ed25519 CAWG4Q 8KzLc949on8iN1pK8q11OpCIeO71t6b0zxCLHhcQ6ns
+-> ssh-ed25519 cK5kHw TP7+OQpQSNuyArnUo1C97J3P3oB0YtzCEPeVvlzsYHE
-uy7z6RdIuoUes+Uap3k5eoFFuu/DcSrEBwq4V4C/ygc
+Bsy5KPNHTVNHnF1sxOvlfJq3CNMVFaXdYkRG2vSj7qM
--> ssh-ed25519 xA739A SLx5cKo0fdAHj+cLpJ4FYTWTUTyDsCqKQOufDu3xnGo
+-> ssh-ed25519 CAWG4Q eQyzwNaH6CfaYIjs8abEuQxt6vxRXsGz69UletMUVDE
-VnS/WsiSaf6RpXuhgfij4pYu4p9hlJl1oXrfYY9rKlQ
+FDcynPO7xg4PWez5Z8gTg5LyE0Wgb3zT9i3Kon67QsU
--> ssh-ed25519 MSF3dg c5ZXvdNxNfZU3HeWsttuhy+UC5JxWN/IFuCuCGbksn4
+-> ssh-ed25519 xA739A 2JuLai2fUu3dZBydS8cMrLrEUIUkz4NNaiupoBOtTwU
-vcKlIirf+VvERX71YpmwW6zp6ClhlG2PR4R8LIN7cQo
+sdM3X+XRzysop7yqa76Z7FAwTHOj91STCtZvfIgCdB0
---- pJKICDaYAlxqNnvHIuzB3Yk7tv0ZNYflGTQD+Zk/8+4
+-> ssh-ed25519 MSF3dg fSPkiWnpInX1V5p3afPCoPotcGFoWFiOMPThtY927lc
-<EFBFBD>h/\J<>J
+8v7E/3l0xA2VWZPXzkN4NmnaA0KJutLMurn/ZXZmhxA
-<EFBFBD>0?<3F> <20>p<EFBFBD><70><EFBFBD>@܉7<DC89><37>3<EFBFBD><33><EFBFBD><EFBFBD>z<EFBFBD><7A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>a<EFBFBD><61>'<27>,ka<6B>I<EFBFBD>XXOZ<4F>I\<5C><><EFBFBD><EFBFBD><EFBFBD>	<09>BP<42><50>/cUɿ~B<><42>S'Q<><51><EFBFBD><EFBFBD>f<06><><EFBFBD>er<65><72><EFBFBD><EFBFBD>^<5E><><EFBFBD><EFBFBD>8l<38><6C>V<EFBFBD>E<EFBFBD><45><EFBFBD>
+--- MQkyBx9hT4ILYXKoZT18PWny1QbDFymcZr63zjMN/qQ
+-b<>#<23><>M.<16>@<40>t<EFBFBD><74><EFBFBD>ŵ}+ό#@<40><><EFBFBD><EFBFBD><EFBFBD>k<EFBFBD>y<EFBFBD><79><EFBFBD>?v<><76>n<1F><>T<EFBFBD>+<2B><><EFBFBD>[<5B>Q<EFBFBD> gA<67><41><EFBFBD>

secrets/vpn-dac-login.age

@@ -1,12 +1,14 @@
 age-encryption.org/v1
--> ssh-ed25519 G5LX5w /RF8uZ/KahUqjEFILbF3+Jin+U0SQdoQChcc9RJ9axc
+-> ssh-ed25519 G5LX5w SRJhNenoQXbT1FgX3TMPnVH5P6oe2eHot+M1YsEjsEk
-aEmPk++86nBR6d2BIa/oaUdyiLS6cH8TUoYJE3bxba4
+hfTSLgKi98Eh7JK5o7x2POpTEtQlQCpEa3keUFYCuME
--> ssh-ed25519 CAWG4Q qHyh9nQi8c3z/KHby9y5vhzN0Dwz0zca98ebjJmXrzs
+-> ssh-ed25519 cK5kHw z5TwWJTkvx7HztjXHJW/aCOtOfPrQaLP0gyIT7rXcyU
-ZbmwNzrSSQ3RvskE8SqcBa0vMy8pzm/HPGHLm5zuPGQ
+b4NCpHfasgvkLLr+6LcWUl60p59aSNnfp3bl2OFYXo0
--> ssh-ed25519 xA739A FlGbfS4bUxA3gVDzb3yPjp4hV8a7aiNBLUctnN3bGEY
+-> ssh-ed25519 CAWG4Q 4VpS1/OnFe8nxcQbRTKNhjsh/ZQ5cbhSMXwK/jjQ+3o
-3fI6SyVjVhh2M8uc/XV3blpdQMPMYi2qzaHNXvx0bvM
+WF9wvOkqVml4UcEzyzeumKuUwCwwr2zvKLMg+PCB8nk
--> ssh-ed25519 MSF3dg 0Bs/aW0nNISS+93It75o6hKZWa7S+LF5bF5ApsJ2fQ8
+-> ssh-ed25519 xA739A 67FhuJ070jBVMt/xbKHWhfri6iIm0FyaFvzQabsvFBM
-y7o0KYDHEen13ndIxg/mYil3eMxxzvYF2pWqhMb+rBU
+1G5/913dDv/r/6p1x/c5YiUnZzrX/LvIj33KW+PN0KU
---- Iqo75G4+02Y9nc1OOkcEx+iQlKnGYCekAx76tRH53wA
+-> ssh-ed25519 MSF3dg Bj/yB4N2wkyHCHC22tcjjJAA4ebSamN0Z4UVX3ZnryI
-<10>
+6D/ZgTs+j+MGDAbPU5zyK0i9zN6tQy68IcOnQZ27mYg
-<EFBFBD>X<EFBFBD><EFBFBD>%f<0C><><12>hX<0B><>R<>c<EFBFBD>+z<><7A>eg<65>
& <20>d<EFBFBD><64><EFBFBD>ק<06><>A<EFBFBD><41><EFBFBD>чXM<58>1<EFBFBD>
+--- 169erk3ICSYLs4FPEuXCn7QlekWhsmSn0Lr+/R14I5Q
+<EFBFBD><EFBFBD><EFBFBD><EFBFBD><05>ҽ3<D2BD>s<EFBFBD>
+w<EFBFBD><EFBFBD>4D<EFBFBD><EFBFBD>b.<2E><><EFBFBD>"|<7C><><EFBFBD>)"<22><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>;<3B>.<2E>ɫ7)<29>LeC<05>=S؟

secrets/wg-fox.age

@@ -0,0 +1,14 @@
+age-encryption.org/v1
+-> ssh-ed25519 cDBabA heyW9/cxgwFX9IexQIXjAQDWGQPNcMXcArQp2Rxsqx4
+o9MQ7EH8PDDjsJdpH9F3Xq2zUoaDAJQlfFmYucSFs6Y
+-> ssh-ed25519 cK5kHw Sza4pos7K3qW3omEeyidI/jszJNf9smemSZnUJfCIww
+D6vazXki7hIYraIuSiGPS+FPbkFUwHhHWDf52OhEIMg
+-> ssh-ed25519 CAWG4Q YexIHueOIMmIN8JIDyNUOKBkyz/k18HqV3hTXh48KlM
+xh8UJzzWT6ByN+Dpn4JrMNsjGC/uc/v6LynwjBDz9NQ
+-> ssh-ed25519 xA739A KySG3TXdqfCMUkVEDGa74B0op745s3XGYxFLyAXSQAc
+5EI/yb5ctW9Qu18bHm3/sK97kwGcKzzmWvPSCWm89XA
+-> ssh-ed25519 MSF3dg MNxnNj0fHmri8ophexXPNjRUBUWrzcuk5S1mucxUMTE
+GVFWXtISEU8ZmlwL4nh4weAgfGrt2GHX0DTzbpS6zg8
+--- UdrqkYG2ZApAuwdZeNhC50NP2rkD/Ol6y8nJa4RHx7Y
+<EFBFBD>ܻ<EFBFBD>m(<28><><EFBFBD>><3E>H<48>Y87<><37>G<0F>+*<12><><EFBFBD><EFBFBD>9V<>.<2E><><EFBFBD><EFBFBD><03><><EFBFBD>p<EFBFBD>Oo<4F>=+哇<>P0<50><30>{<7B>)<29><17><><EFBFBD><EFBFBD>><3E>z3P^
+u

web/content/fox/_index.md

@@ -21,17 +21,28 @@ the detailed specifications:
 
 ## Access
 
-To access the machine, request a SLURM session from [hut](/hut) using the `fox`
+To access the machine, request a SLURM session from [apex](/apex) using the `fox`
-partition:
+partition. If you need the machine for performance measurements, use an
+exclusive reservation:
 
-    hut% salloc -p fox
+    apex% salloc -p fox --exclusive
 
-Then connect via ssh:
+Otherwise, specify the CPUs that you need so other users can also use the node
+at the same time:
 
-    hut% ssh fox
+    apex% salloc -p fox -c 8
+
+Then use srun to execute an interactive shell:
+
+    apex% srun --pty $SHELL
     fox%
 
-Follow [these steps](/access) if you don't have access to hut or fox.
+Make sure you get all CPUs you expect:
+
+    fox% grep Cpus_allowed_list /proc/self/status
+    Cpus_allowed_list:	0-191
+
+Follow [these steps](/access) if you don't have access to apex or fox.
 
 ## CUDA
 
@@ -89,9 +100,8 @@ Then just run `nix develop` from the same directory:
 
 The machine has several file systems available.
 
-- `$HOME`: Mounted via NFS across all nodes. It is slow and has low capacity.
+- `/nfs/home`: The `/home` from apex via NFS, which is also shared with other
-  Don't abuse.
+  xeon machines. It has about 2 ms of latency, so not suitable for quick random
-- `/ceph/home/$USER`: Shared Ceph file system across jungle nodes. Slow but high
+  access.
-  capacity. Stores three redundant copies of every file.
 - `/nvme{0,1}/$USER`: The two local NVME disks, very fast and large capacity.
 - `/tmp`: tmpfs, fast but not backed by a disk. Will be erased on reboot.